Goldman exec warns bankers risk ‘cognitive atrophy’ from overusing AI models: ‘We’re delegating reasoning’

“There’s a huge danger here that in the era of AI, we outsource our reasoning to these models, and we have cognitive atrophy that stops us being able to reason from first principles ourselves,” Churchman said, according to a transcript provided to CNBC.

His point is not that AI has no place on Wall Street. It is that, much like GPS made us lose some of our navigation knowledge and smartphones made memorization less necessary, relying too heavily on algorithms could gradually dull the analytical skills bankers are expected to bring to the job in the first place.

The hidden cost of efficiency

A company’s embrace of AI comes with an appealing short-term promise: greater efficiency, lower costs and potentially higher profits. The longer-term trade-off is more complicated. In automating the work traditionally handed to junior bankers and traders, firms may also be chipping away at the apprenticeship system that teaches them how to become senior ones.

That pressure is already beginning to show up in the job market. Employment among workers ages 22 to 25 in highly AI-exposed occupations is now 19% below where it would be if it had kept pace with young workers in less-exposed jobs, according to Stanford’s Digital Economy Lab. Researchers found the gap was driven largely by companies hiring fewer young workers, rather than laying off those already on the job.

Seb Kirk, CEO and co-founder of AI solutions platform GaiaLens, said the danger is not simply that young workers will miss out on tedious assignments. It is that they may lose the exposure that came with doing them.

“Nobody ever learned judgment from reformatting a pitch deck at two in the morning,” Kirk told Moneywise. “What you got out of that work was exposure. You were handling the numbers yourself, so you noticed when one of them looked wrong. The tedium and the learning happened to be bundled into the same job, and the industry is now confusing the two.”

For Wall Street, that creates an awkward paradox. The same technology that can make firms leaner and more profitable today could also leave them with fewer people developing the skills needed to supervise those systems tomorrow.

“The bit that worries me more is the arithmetic,” he said. “Thin out the junior ranks and in ten years you have fewer people capable of spotting when the model is confidently wrong.”

Finding the balance

For Churchman, the debate is no longer about whether people will use AI. It is about how much of the job firms should hand over to it.

And the transition is happening quickly. According to Gallup, 30% of U.S. employees said they were using AI at work at least a few times a week, while 15% were using it daily.

Churchman has seen firsthand how that shift could play out on a trading floor. Before joining Goldman Sachs in 2021, he ran currency trading at UBS, where junior traders learned the business in part by handling client pricing requests under the supervision of more experienced colleagues. Much of that work can now be automated.

“We can absolutely automate that,” he said, “but then do we get the senior traders that fully understand?”

Keeping humans in the loop

That is the balance many companies are trying to strike: using AI to remove repetitive work without removing the judgment-building moments that come with it. Churchman said humans still need to remain responsible for decisions where the stakes are high and the outcome is uncertain. Even Goldman, he acknowledged, has not yet fully figured out what that model should look like.